The Status Quo And Prediction Of Research Collaboration In The Field Of Biomedine

PurposeBy combining theories with empirical studies, as well as qualitative methods withquantitative methods, we aimed to apply scientometrics theories to the evaluation ofthe research collaboration in biomedical fields in the background of knowledgenetwork. We sorted out the traditional research collaboration theory system and madeattempts to establish a new one–the theory of collaboration relationship prediction.Then we confirmed the validity of this system through an empirical study. Theseefforts will help the research management departments of various levels andresearchers have a better understanding of the structure of China’s innovation systemand get a right picture of the importance and necessity of research collaboration. Webelieve this study will provide scientific evidences and suggestions for researchmanagement policymaking.Research subjectsThree subfields of biomedicine were chosen for this study: Cardiology&Cardiovasology, Oncology, Health Management.Data1. Data in the fields of Cardiology&Cardiovasology (C&C)(1) Chinese dataFive major journals in C&C field were chosen from ‘A Guide to the CoreJournals of China (2008Edition)’. Then, the bibliographic records published in these 5journals from the year2000through the year2010were collected from databases inChina National Knowledge Infrastructure (CNKI) and VIP Journal IntegrationPlatform (VIP).(2) International dataa. All the bibliographic records containing ‘‘coronary’ in their title, abstract orkeywords from the year1981through the year2010were collected from ScienceCitation Index-expanded (SCI-E) in Web of Science (WoS).b. Top four journals with5-year Journal Impact Factor greater than10werechosen from category “Cardiac&Cardiovascular System” in Journal Citation Report.Then, all the bibliographic records published in these journals from the year2001tothe year2010were collected from SCI-E.2. Data in the fields of oncology(1) Chinese dataWe chose all the10Core Oncology journals from Chinese Science CitationDatabase. Then, all the bibliographic records published in these10journals from theyear2000through the year2009were collected from databases in CNKI and VIP.(2) International dataTop10%(30) journals in terms of both Impact Factor and Citations were chosenfrom category “Oncology” in Journal Citation Report. Then, all the bibliographicrecords in these journals from the2001to the2010were collected from SCI-E,CPCI-SSH of Web of Science database.3. Data in the fields of health management(1) Chinese dataAll the bibliographic records in “Chinese Journal of Health Management” from2007-2012were collected from VJIP, Wangfang Data and CD-ROM version ofChinese Journal of Health Management.(2) International dataAll the bibliographic records with the keyword “health management” from theyear1999to the year2011were collected from WoS.Methods 1. Literature analysis.2. Scientomerics: Co-authorship and Co-word analysis.3. Social network analysis: Component analysis, Centrality analysis, K-coreanalysis, M-slice analysis and Clique analysis.4. Statistical methods: Mann–Whitney U test, Binary logistic regression, Orderedlogistic regression, Hierarchical Clustering and Multidimensional Scaling.5. Machine learning: Logistic regression (LR), Support Vector Machine (SVM),Lpmade and WEKA.6. Programming language: Java and Python; Database management tool: MySQL.7. Empirical study.Results1. The status quo of research collaboration in biomedical domain(1) The fields of C&CChina: From the year2000to the year2010, the percentage of co-authoredpapers and the average number of authors per paper in Chinese C&C field weregenerally increasing. The geographic distribution of the research collaborationactivity was extremely uneven.87authors and5institutions ranked in top1%of allthe three centralities.92.8%authors belonged to10-Core and below.90.93%authorsbelonged to3-slice and below.63cohesive research groups were found. Coronaryartery disease, myocardial infarction, etc. were the focuses of research collaboration.The world: From the year1981to the year2010, research collaborations hadincreased at the author, institution and countries/regions level in international CHDresearch.3000most collaborative authors,572most collaborative institutions and52countries/regions were extracted from their corresponding collaboration network.766Cliques were found in the most collaborative authors.308Cliques were found in themost collaborative institutions. Western countries/regions represented the core of theworld’s collaboration, while eastern countries/regions scattered at the periphery of thenetwork. The rate of economic development in the countries/regions affected themulti-national collaboration behavior as well. (2) The fields of OncologyChina: From the year2000to the year2009, the research collaboration In thefield of Chinese oncology research were mostly distributed in eastern regions such asBeijing, Shanghai and Guangdong. Three Centrality measures correlated with therankings of author’s productivity. Most authors (90%) belonged to small K-Core(smaller than9).92.25%of all the authors belonged to M-Slices lower than four.480authors were in11-slice and above, about1%of all the authors.12groups weregenerated by using hierarchical clustering analysis.6groups were found by usingmultidimensional scaling analysis.The world: From the year2001to the year2010,36most collaborative academiccommunities were indentified in international oncology research domain. NakamuraY was the most productive author, publishing117papers. The largest academiccommunity contains8authors who were mainly from the Harvard University Schoolof Medicine, the Division of Hematology and Oncology of the Dana Farber CancerInstitute, and the Jerome Lipper Multiple Myeloma Center of Boston. All the5authors in the second Component were affiliated with the Department ofNeurosurgery of the University of Illinois College of Medicine at Peoria. All the4authors in the third Component were from the Department of Leukemia of the MDAnderson Cancer Center at the University of Texas. The primary focuses of these36academic communities were multiple myeloma, angiogenesis and lymphocyticleukaemia, etc.(3) The fields of Health ManagementChina: From the year2007to the year2012,1933authors and625institutionswere involved in the health management research activities. The average number ofauthors per paper was3. The number of authors with papers no less than4orcentrality larger than0was54. Hunang Jianshi, Wu Liuxin and Zeng Qiang toped theranking of productivity and centrality, and together with Bai Shuzhong, Tian Jingfaand Han Jing, they formed a broader collaboration network. Wang Peiyu and Du Bingformed their own collaboration groups as well. High productivity authors weremostly from Air Force Institute of Aeromedicin, Peking University Health Science Center, Beijing Physical Examination Center and Chinese Academy of MedicalScience&Peking Union Medical College.The world: From the year1999to the year2011, both the productions andresearch collaboration at the author, institution and countries/regions levels wereincreasing in health managment research domain.17researchers (O’Toole T1, etc)can be seen as the academic leaders in this field.37research institutions (Universityof Minnesota, etc) played a vital role in the information dissemination and resourcescontrol in health management. The Component analysis found that22research groupscan be regarded as the backbone in this field. The8institution groups consisting of33institutions formed the core of this field. USA, UK and Australia lied in the center bycohesive subgroup analysis. High frequency keywords such as care, disease, systemand model were involved in the health management field.2. The relationship between research collaboration and research performance inbiomedical domainThe relationship between research performance of scientists and internationalcollaboration: In the field of C&C, both international publications andinter-institutional publications were higher quality than those published by scientistswithin the same institution, and that there was a significant positive correlationbetween scientists’research performance and international collaboration.The relationship between research performance of countries and internationalcollaboration: In international Coronary Heart Disease (CHD) research domain, themost productive countries were USA, Japan, German, British, etc. The mostinfluential countries included USA, British and Germain. However, Hungary,Switzerland, Danmark, Austria, Finland, and Norway led in the ranking measured bytheir proportion of collaborative output.3. The research collaboration prediction in biomedical domainThe collaboration prediction in CHD research showed that: Both LR model andSVM model scored well for all the four evaluation measures. SVM model beated LRmodel in terms of3evaluation measures: precision rate, recall rate and AUC. Both learning models generally scored high for high productivity author sets in terms of allthe four evaluation measures, but scored low for less productivity author sets. Wetrained the two models with the selected features on the entire author set, and foundthat the testing results were improved for both the LR model and SVM model. TheLR model generally produced relatively lower accuracy rates when testingtopological features separately than it did when testing all the topological features asa whole.Conclusions1. The research collaboration in biomedical domain were unevenly distributedThe tendency to work in group among researchers in producing scientificpublications in Chinese C&C field was increasing over time. Collaboration amongscientists, institutios and countries in CHD research had significantly grown over thepast three decades.However, the research collaboration activities in diffentent regions of China inC&C fields were unevenly distributed, and western countries were placing at the coreof the international CHD research collaboration network. Moreover, the primarycollaborators of high incomes countries were high incomes ones, then the middleincomes and low incomes ones.2. Centrality analysis can help to choose the discipline academic leaderDegree centrality, Closeness centrality and Betweenness centrality can be goodindicators of author’s productivity, then help to indentify the academic leaders inbiomedical research domain.3. Cohesive group analysis can help to extract best research groups in biomedicaldomainCohesive group analysis such as Component analysis, K-core analysis, M-sliceanalysis and Clique analyis can be used to extract the outstanding research groups inthe biomedical research domain. 4. There was a significant positive correlation between scientists’ researchperformance and international collaborationThe quality of research papers was influenced by the degree of reserachcollaboration, and there was a significant positive correlation between scientists’research performance and international collaboration.5. The topological features in co-authorship network can be used to makecollaboration relationship predictionBoth the traditionally used algorithm LR and increasingly promising algorithmSVM model performed well in co-author relationship prediction. The collaborationrelationships for high productive authors were easier to predict than less productiveauthors in terms of all the four evaluation measures. Testing results for both modelswere improved through feature selection. The co-author relationship predictionshould be made by using these topological features as a whole instead of using asingle one.Suggestions1. Promoting the research performance of scientists by strengthening the researchcollaborationThe research quantity and quality of scientists in the field of biomedicine can bepromoted by strengthening the research collaboration. Results showed that theresearch collaboration degree in biomedical rearsch domain were increasing, howeverthe research collaborations were unevenly distributed. So we should encouragecross-regional research collaboration and international research collaboration inbiomedical research domain so that the biomedical research can develop in harmony.2. Social network analysis should be applied to the scholarly network analysisBy applying social network analysis (SNA) to the coauthorship analysis,centrality analysis helped to choose the discipline academic leader, and cohesivegroup analysis helped to extract outstanding research groups. This again proved the applicability and feasibility of SNA in the field of scientometrics.3. Data mining should be a good complement to scientometricsWe made an initial attempt to use data mining techniques in this study. Wepresented supervised machine learning methods for building link prediction modelsfrom topological features of node pairs in co-authorship networks. The models couldbe useful in identifying unrealized yet potentially sucessful collaboration, whichwould in turn faciliate the development of strong research groups, and provideevidence and guiding for research management and policy-making.Innovation points1. Developing a brand-new theory for research collaboration--The prediction ofresearch collaboration relationshipThe theory of research collaboration evolution and prediction, together withtraditional research collaboration theories, composed a complete theory system ofresearch collaboration. The solving of research collaboration evolution and predictionproblem can help to organize and manage strong research team, and improve thework efficiency.2. Applying data mining, machine learning social network analysis, multivariatestatistics and computer programming to scientometricsData mining, machine learning, prediction analysis, social network analysis, andmultivariate statistics were combined with scientometrics methods to explore theresearch collaboration in biomedical domain, and Java, Python and MySQL wereused to convert CSV, TSV and XML files into the data input for WEKA, Lpmade,MySQL, etc.3. The results of extracting highly influential scientists and research groups andco-author relationship link prediction can provide scientific evidences for China’sresearch management policymakingThis study proved the significant positive correlation between scientists’ research performance and international collaboration, verified the excellent researchperformance of the research groups found by cohesive groups analysis, and confirmedthe practicability of the co-author relationship prediction by using topoligical featuresin co-authorship network. These facts can provide scientific evidences for China’sresearch management policymaking.Conceptions for making further study1. Research collaboration prediction in heterogeneous networkWe will examine the collaboration predictions in heterogeneous network basedon the results of link prediction in homogeneous network. As an heterogeneousnetwork contain multiply types of nodes and edges, it can provide more topologicalfeatures, and make the link prediction more accurately.2. The combination of Topic modeling with co-authorship networkTopic modeling has been widely used in the field of natural language processing.We will apply the LDA method to the co-authorship network so that the “researchsubject” can be incorporated into the collaboration analysis more properly.The dissertation is done as part of the project ‘‘Cooperation Analysis ofTechnology Innovation Team Member Based on Knowledge Network-EmpiricalEvidence in the Biology and Biomedicine Field (No.71103114)’’ and the project‘‘Scientific and Technological Collaboration in the Field of Biomedicine-UsingCo-authorship and Co-inventorship Analysis (No.71240006)’’, both supported byNational Natural Science Foundation of China.